2026-09-09 –, Ballroom All times in US/Pacific
Your on-call process was built for software bugs. A service goes down, an alert fires, someone pages the right engineer, and you work the problem. But AI systems fail differently: they degrade quietly, hallucinate confidently, and resist the clean root cause analysis your runbooks were written around. If 2025 was the year of AI adoption, 2026 is shaping up to be the year everyone finds out their incident response process wasn't built for this.
This talk examines what's genuinely new about AI incidents: the failure modes, the metrics that matter, and why your existing playbooks miss them entirely. We'll dig into what needs to change across process, roles, and communication, and what teams handling this well are already doing differently, before the next AI failure exposes the gap.
Kat spent years leading technical support teams on the front lines of incident response before moving to the stage to talk about it. As Senior Manager of Developer Relations at PagerDuty, she leverages 12+ years of experience to help teams communicate clearly when everything is on fire. She speaks at conferences globally on sustainable incident management practices, cross-functional collaboration, and career development. When she's not wrangling incidents, she's wrangling two rabbits (Lupin and Ginny), a dog named Violet, and an ambitious collection of houseplants.